servediff 0.1.0__py3-none-any.whl

servediff/__init__.py

@@ -0,0 +1,6 @@
+"""servediff: diff what an LLM inference server actually ran with."""
+
+from servediff.cli import diff, main, parse
+
+__version__ = "0.1.0"
+__all__ = ["diff", "main", "parse", "__version__"]

servediff/__main__.py

@@ -0,0 +1,5 @@
+import sys
+
+from servediff.cli import main
+
+sys.exit(main())

servediff/cli.py

@@ -0,0 +1,302 @@
+"""servediff: what did the inference server actually run with? Diff two server logs.
+
+Benchmark result files record the client side only. The server's effective
+configuration -- auto-selected attention/MoE backends, CUDA graph and compile
+mode, the arguments it resolved -- is only printed in its startup log. This tool
+extracts those facts from vLLM and SGLang logs and diffs two of them.
+
+Every fact keeps the line it came from. A fact that is not in the log is reported
+as absent, never filled with a default.
+
+Usage:
+    servediff show <log>
+    servediff diff <baseline.log> <candidate.log> [--raw-args] [--json] [--all]
+    servediff env [--python /path/to/server/venv/bin/python] [-o env.json]
+    servediff diff <baseline-env.json> <candidate-env.json>
+Exit codes: 0 no differences, 1 differences found, 2 nothing recognised in an input.
+"""
+
+from __future__ import annotations
+
+import argparse
+import ast
+import json
+import re
+import sys
+from dataclasses import dataclass, field
+from pathlib import Path
+
+from servediff import env as envsnap
+
+ANSI = re.compile(r"\x1b\[[0-9;]*m")
+
+
+@dataclass
+class Fact:
+    values: list = field(default_factory=list)  # distinct values in order seen
+    evidence: list = field(default_factory=list)  # (line number, line text)
+
+    def add(self, value, lineno: int, line: str) -> None:
+        if value not in self.values:
+            self.values.append(value)
+            self.evidence.append((lineno, line.strip()[:300]))
+
+    @property
+    def value(self):
+        return self.values[0] if len(self.values) == 1 else self.values
+
+
+# Patterns over one log line. Each yields (fact key, value) pairs.
+LINE_RULES: list[tuple[re.Pattern, callable]] = [
+    # vLLM banner and engine init.
+    (re.compile(r"\bversion (\d+\.\d+\.\d+\S*)"), lambda m: [("engine.version", m[1])]),
+    (re.compile(r"Initializing a V1 LLM engine \(v([^)]+)\)"), lambda m: [("engine.version", m[1])]),
+    # vLLM: "Using FLASH_ATTN attention backend out of potential backends: [...]".
+    (re.compile(r"Using (\S+) attention backend"), lambda m: [("backend.attention", m[1])]),
+    (re.compile(r"Using FlashAttention version (\d+)"), lambda m: [("backend.flash_attn_version", m[1])]),
+    # vLLM MoE selection, three formats seen in the wild:
+    #   Using 'VLLM_CUTLASS' NvFp4 MoE backend
+    #   Using FlashInfer CUTLASS Unquantized MoE backend out of potential backends: [...]
+    #   Using TrtLlmBf16ExpertsMonolithic MoE backend
+    (re.compile(r"Using '?([^']+?)'? (?:(NvFp4|Fp8|Unquantized|Mxfp4|Int4|W4A16) )?MoE backend"),
+     lambda m: [("backend.moe", m[1] if m[2] is None else f"{m[1]} ({m[2]})")]),
+    (re.compile(r"Using (MoEPrepareAndFinalize\w+)"), lambda m: [("backend.moe_prepare_finalize", m[1])]),
+    (re.compile(r"Enforce eager set"), lambda m: [("cuda_graph", "disabled (enforce_eager)"),
+                                                  ("compile", "disabled (enforce_eager)")]),
+    (re.compile(r"cudagraph_mode[=:] ?'?(\w+)"), lambda m: [("cuda_graph.mode", m[1])]),
+    (re.compile(r"Enabled custom fusions: (.+)$"), lambda m: [("compile.fusions", m[1].strip())]),
+    # tqdm bars: "Capturing CUDA graphs (decode, FULL):   0%| 0/35"; one value per mode and size.
+    (re.compile(r"Capturing CUDA graphs \(([^)]+)\):\s+\d+%\|[^|]*\|\s*\d+/(\d+)"),
+     lambda m: [("cuda_graph.captures", f"{m[1]} x{m[2]}")]),
+    (re.compile(r"CUDA graph pool memory: ([\d.]+) GiB \(actual\)"),
+     lambda m: [("cuda_graph.pool_gib", float(m[1]))]),
+    (re.compile(r"FlashInfer top-p/top-k sampling (disabled|enabled)"),
+     lambda m: [("backend.sampling_flashinfer", m[1])]),
+    (re.compile(r"Using (V\d) Model Runner"), lambda m: [("model_runner", m[1])]),
+    (re.compile(r"Pinned memory is not available.*falling back to (.+?) for UVA buffers"),
+     lambda m: [("uva_buffers", f"fallback: {m[1]}")]),
+    # Startup outcome. Without it, facts missing after a crash look like "not printed".
+    (re.compile(r"Application startup complete"), lambda m: [("startup", "ready")]),
+    (re.compile(r"EngineCore failed to start|Engine core initialization failed"),
+     lambda m: [("startup", "failed")]),
+    (re.compile(r"^(?:\(\w+ pid=\d+\) )?(?:ERROR .*?\] )?((?:RuntimeError|ValueError|torch\.OutOfMemoryError|AssertionError): .+)$"),
+     lambda m: [("startup.error", m[1].strip())]),
+    # SGLang.
+    (re.compile(r"Disable prefill CUDA graph because"), lambda m: [("cuda_graph.prefill", "disabled")]),
+    (re.compile(r"The server is fired up and ready to roll"), lambda m: [("startup", "ready")]),
+    (re.compile(r"KV Cache is allocated\. dtype: (\S+?),"), lambda m: [("kv_cache.dtype", m[1])]),
+]
+
+# Arguments that mean the same thing in both engines, mapped to one name.
+# value: (vLLM non-default arg, SGLang server_arg, transform for SGLang value)
+SHARED_ARGS = {
+    "model": ("model", "model_path", None),
+    "dtype": ("dtype", "dtype", None),
+    "max_model_len": ("max_model_len", "context_length", None),
+    "gpu_memory_fraction": ("gpu_memory_utilization", "mem_fraction_static", None),
+    "prefix_cache": ("enable_prefix_caching", "disable_radix_cache", lambda v: not v),
+    "cuda_graph.disabled": ("enforce_eager", "disable_cuda_graph", None),
+    "tensor_parallel_size": ("tensor_parallel_size", "tp_size", None),
+    "kv_cache_dtype": ("kv_cache_dtype", "kv_cache_dtype", None),
+    "backend.attention_arg": ("attention_backend", "attention_backend", None),
+    "backend.moe_arg": ("moe_backend", "moe_runner_backend", None),
+    "backend.sampling_arg": (None, "sampling_backend", None),
+}
+
+
+def _literal_dict(text: str) -> dict | None:
+    """Parse the `{...}` that starts `text`, tolerating trailing log text."""
+    depth, quote = 0, None
+    for i, ch in enumerate(text):
+        if quote:
+            if ch == quote and text[i - 1] != "\\":
+                quote = None
+        elif ch in "'\"":
+            quote = ch
+        elif ch in "{[(":
+            depth += 1
+        elif ch in "}])":
+            depth -= 1
+            if depth == 0:
+                try:
+                    value = ast.literal_eval(text[: i + 1])
+                except (ValueError, SyntaxError):
+                    return None
+                return value if isinstance(value, dict) else None
+    return None
+
+
+def parse(text: str) -> dict:
+    facts: dict[str, Fact] = {}
+    args: dict = {}
+    engine = None
+
+    def put(key, value, lineno, line):
+        facts.setdefault(key, Fact()).add(value, lineno, line)
+
+    for lineno, raw in enumerate(text.splitlines(), 1):
+        line = ANSI.sub("", raw)
+        if "non-default args:" in line:
+            engine = engine or "vllm"
+            parsed = _literal_dict(line.split("non-default args:", 1)[1].strip())
+            if parsed is not None:
+                args.update(parsed)
+                put("args.source", "vllm non-default args", lineno, line)
+        elif "server_args=" in line:
+            engine = engine or "sglang"
+            parsed = _literal_dict(line.split("server_args=", 1)[1].strip())
+            if parsed is not None:
+                args.update(parsed)
+                put("args.source", "sglang server_args (all args)", lineno, line)
+        if engine is None and ("vllm" in line.lower() and "[" in line and ".py:" in line):
+            engine = "vllm"
+        for pattern, emit in LINE_RULES:
+            match = pattern.search(line)
+            if match:
+                for key, value in emit(match):
+                    # Errors after the server reported ready (e.g. during shutdown) say
+                    # nothing about how it started.
+                    if key == "startup.error" and "ready" in (facts.get("startup") or Fact()).values:
+                        key = "after_ready.error"
+                    put(key, value, lineno, line)
+
+    if engine:
+        facts.setdefault("engine", Fact()).add(engine, 0, "(detected from log format)")
+    for name, (vllm_key, sglang_key, transform) in SHARED_ARGS.items():
+        key = vllm_key if engine == "vllm" else sglang_key
+        if key and key in args:
+            value = args[key]
+            if engine == "sglang" and transform:
+                value = transform(value)
+            if isinstance(value, list) and len(value) == 1:
+                value = value[0]
+            put(f"arg.{name}", value, 0, f"{'non-default args' if engine == 'vllm' else 'server_args'}[{key!r}]")
+    return {"engine": engine, "facts": facts, "args": args}
+
+
+def _shown(fact: Fact | None):
+    return "<absent>" if fact is None else fact.value
+
+
+def diff(a: dict, b: dict, raw_args: bool = False) -> list[dict]:
+    rows = []
+    for key in sorted(set(a["facts"]) | set(b["facts"])):
+        fa, fb = a["facts"].get(key), b["facts"].get(key)
+        va, vb = _shown(fa), _shown(fb)
+        if fa is None or fb is None:
+            status = "only_baseline" if fb is None else "only_candidate"
+        else:
+            status = "same" if va == vb else "changed"
+        rows.append({"key": key, "baseline": va, "candidate": vb, "status": status,
+                     "evidence": {"baseline": fa.evidence if fa else [],
+                                  "candidate": fb.evidence if fb else []}})
+    if raw_args and a["engine"] == b["engine"]:
+        for key in sorted(set(a["args"]) | set(b["args"])):
+            va, vb = a["args"].get(key, "<absent>"), b["args"].get(key, "<absent>")
+            if va != vb:
+                rows.append({"key": f"raw_arg.{key}", "baseline": va, "candidate": vb,
+                             "status": "changed", "evidence": {}})
+    return rows
+
+
+def _fmt(value) -> str:
+    text = json.dumps(value, ensure_ascii=False) if not isinstance(value, str) else value
+    return text if len(text) <= 60 else text[:57] + "..."
+
+
+def main(argv=None) -> int:
+    parser = argparse.ArgumentParser(description=__doc__.split("\n\n")[0])
+    sub = parser.add_subparsers(dest="command", required=True)
+    show = sub.add_parser("show", help="effective config facts in one log")
+    show.add_argument("log", type=Path)
+    show.add_argument("--json", action="store_true")
+    d = sub.add_parser("diff", help="diff two server logs, or two `servediff env` snapshots")
+    d.add_argument("baseline", type=Path)
+    d.add_argument("candidate", type=Path)
+    d.add_argument("--raw-args", action="store_true", help="also diff every raw argument (same engine only)")
+    d.add_argument("--json", action="store_true")
+    d.add_argument("--all", action="store_true", help="also print unchanged facts")
+    e = sub.add_parser("env", help="snapshot package versions, GPU and engine env vars of a server environment")
+    e.add_argument("--python", help="interpreter of the server's environment (default: this one)")
+    e.add_argument("-o", "--output", type=Path, help="write JSON here instead of stdout")
+    args = parser.parse_args(argv)
+
+    if args.command == "env":
+        try:
+            snap = envsnap.snapshot(args.python)
+        except (RuntimeError, OSError) as exc:
+            print(f"servediff env: {exc}", file=sys.stderr)
+            return 2
+        text = json.dumps(snap, indent=2, ensure_ascii=False, sort_keys=True)
+        if args.output:
+            args.output.write_text(text + "\n", encoding="utf-8")
+            print(f"wrote {args.output}: {len(snap['packages'])} packages, {len(snap['env'])} env vars", file=sys.stderr)
+        else:
+            print(text)
+        return 0
+
+    if args.command == "show":
+        parsed = parse(args.log.read_text(encoding="utf-8", errors="replace"))
+        if not parsed["facts"]:
+            print(f"{args.log}: no vLLM or SGLang configuration lines recognised", file=sys.stderr)
+            return 2
+        if args.json:
+            print(json.dumps({k: {"value": f.value, "evidence": f.evidence} for k, f in parsed["facts"].items()},
+                             indent=2, ensure_ascii=False))
+        else:
+            for key, fact in sorted(parsed["facts"].items()):
+                where = fact.evidence[0][0]
+                print(f"{key:32s} {_fmt(fact.value):60s} line {where if where else '-'}")
+        return 0
+
+    text_a = args.baseline.read_text(encoding="utf-8", errors="replace")
+    text_b = args.candidate.read_text(encoding="utf-8", errors="replace")
+    snaps = envsnap.is_snapshot(text_a), envsnap.is_snapshot(text_b)
+    if any(snaps):
+        if not all(snaps):
+            print("diff: give two server logs or two `servediff env` snapshots, not one of each", file=sys.stderr)
+            return 2
+        rows = envsnap.ranked(envsnap.diff(json.loads(text_a), json.loads(text_b)))
+        if args.json:
+            print(json.dumps({"differences": len(rows), "rows": rows}, indent=2, ensure_ascii=False))
+        else:
+            key = [r for r in rows if envsnap.is_key(r)]
+            shown = rows if args.all else key
+            print(f"{'fact':32s} {'baseline':40s} {'candidate':40s} status")
+            for r in shown:
+                print(f"{r['key']:32s} {_fmt(r['baseline'])[:40]:40s} {_fmt(r['candidate'])[:40]:40s} {r['status']}")
+            print(f"\n{len(key)} difference(s) in Python, GPU, inference packages and engine env vars;"
+                  f" {len(rows) - len(key)} in other packages{'' if args.all else ' (--all to list them)'}.")
+        return 1 if rows else 0
+
+    a = parse(text_a)
+    b = parse(text_b)
+    for path, parsed in ((args.baseline, a), (args.candidate, b)):
+        if not parsed["facts"]:
+            print(f"{path}: no vLLM or SGLang configuration lines recognised", file=sys.stderr)
+            return 2
+    rows = diff(a, b, raw_args=args.raw_args)
+    if not args.all:  # errors after "ready" (e.g. on shutdown) say nothing about the config
+        rows = [r for r in rows if not r["key"].startswith("after_ready.")]
+    differing = [r for r in rows if r["status"] != "same"]
+    if args.json:
+        print(json.dumps({"differences": len(differing), "rows": rows}, indent=2, ensure_ascii=False, default=str))
+    else:
+        if a["engine"] != b["engine"]:
+            print(f"note: comparing {a['engine']} with {b['engine']}; only shared facts line up\n")
+        for side, parsed in (("baseline", a), ("candidate", b)):
+            if "failed" in (parsed["facts"].get("startup") or Fact()).values:
+                error = parsed["facts"].get("startup.error")
+                print(f"warning: {side} server failed to start"
+                      f"{': ' + _fmt(error.values[0]) if error else ''}; facts after the failure"  # first = root cause
+                      " are absent because startup stopped, not because they were off\n")
+        print(f"{'fact':32s} {'baseline':40s} {'candidate':40s} status")
+        for r in rows if args.all else differing:
+            print(f"{r['key']:32s} {_fmt(r['baseline'])[:40]:40s} {_fmt(r['candidate'])[:40]:40s} {r['status']}")
+        print(f"\n{len(differing)} difference(s); {len(rows) - len(differing)} fact(s) identical."
+              " Absent means the log did not print it, not that it was off.")
+    return 1 if differing else 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

servediff/env.py

@@ -0,0 +1,111 @@
+"""Snapshot the environment a server runs in: package versions, GPU, relevant env vars.
+
+Logs do not show dependency versions (torch, triton, flashinfer...), yet upgrades of
+those change server behaviour. `snapshot()` runs a small stdlib-only probe under the
+server's own interpreter, so nothing has to be installed into that environment.
+"""
+
+from __future__ import annotations
+
+import json
+import re
+import shutil
+import subprocess
+import sys
+import tempfile
+
+SCHEMA = "servediff-env/1"
+
+# Variables that steer inference engines. Values of names that look secret are redacted.
+ENV_PREFIXES = ("VLLM_", "SGLANG_", "SGL_", "CUDA_", "NCCL_", "TORCH_", "PYTORCH_", "FLASHINFER_",
+                "TRITON_", "HF_HUB_OFFLINE", "TRANSFORMERS_OFFLINE")
+SECRET = re.compile(r"TOKEN|KEY|SECRET|PASSWORD|PASSWD|CREDENTIAL|AUTH", re.I)
+
+# Runs under the target interpreter: standard library only.
+PROBE = r"""
+import importlib.metadata as m, json, os, platform, sys
+prefixes = tuple(json.loads(sys.argv[1])); secret = __import__("re").compile(sys.argv[2], 2)
+pkgs = {}
+for d in m.distributions():
+    name = (d.metadata["Name"] or "").lower().replace("_", "-")
+    if name:
+        pkgs[name] = d.version
+env = {k: ("<redacted>" if secret.search(k) else v) for k, v in os.environ.items() if k.startswith(prefixes)}
+print(json.dumps({"python": platform.python_version(), "executable": sys.executable,
+                  "platform": platform.platform(), "packages": pkgs, "env": env}))
+"""
+
+
+def _gpus() -> list[dict] | None:
+    if not shutil.which("nvidia-smi"):
+        return None
+    try:
+        out = subprocess.run(
+            ["nvidia-smi", "--query-gpu=name,driver_version,memory.total", "--format=csv,noheader,nounits"],
+            capture_output=True, text=True, timeout=20, check=True,
+        ).stdout
+    except (subprocess.SubprocessError, OSError):
+        return None
+    gpus = []
+    for line in out.strip().splitlines():
+        name, driver, memory = (part.strip() for part in line.split(","))
+        gpus.append({"name": name, "driver": driver, "memory_mib": int(float(memory))})
+    return gpus
+
+
+def snapshot(python: str | None = None) -> dict:
+    # Not `-I`: it hides user site-packages, which the server itself would see.
+    # A neutral cwd keeps the caller's directory out of sys.path.
+    result = subprocess.run(
+        [python or sys.executable, "-c", PROBE, json.dumps(ENV_PREFIXES), SECRET.pattern],
+        capture_output=True, text=True, timeout=120, cwd=tempfile.gettempdir(),
+    )
+    if result.returncode != 0:
+        raise RuntimeError(f"environment probe failed under {python or sys.executable}: {result.stderr.strip()[-500:]}")
+    data = json.loads(result.stdout)
+    data["schema"] = SCHEMA
+    data["gpus"] = _gpus()
+    return data
+
+
+def is_snapshot(text: str) -> bool:
+    try:
+        return json.loads(text).get("schema", "").startswith("servediff-env/")
+    except (ValueError, AttributeError):
+        return False
+
+
+def diff(a: dict, b: dict) -> list[dict]:
+    rows = []
+
+    def add(key, va, vb):
+        if va != vb:
+            status = "only_candidate" if va is None else "only_baseline" if vb is None else "changed"
+            rows.append({"key": key, "baseline": "<absent>" if va is None else va,
+                         "candidate": "<absent>" if vb is None else vb, "status": status})
+
+    add("python", a.get("python"), b.get("python"))
+    ga = [(g["name"], g["driver"]) for g in a.get("gpus") or []]
+    gb = [(g["name"], g["driver"]) for g in b.get("gpus") or []]
+    add("gpus", ga or None, gb or None)
+    for name in sorted(set(a["packages"]) | set(b["packages"])):
+        add(f"pkg.{name}", a["packages"].get(name), b["packages"].get(name))
+    for name in sorted(set(a["env"]) | set(b["env"])):
+        add(f"env.{name}", a["env"].get(name), b["env"].get(name))
+    return rows
+
+
+# Packages whose version is most likely to change inference behaviour; shown first.
+KEY_PACKAGES = re.compile(r"^pkg\.(vllm|sglang|sgl-kernel|torch|triton|flashinfer.*|xformers|deep-gemm|transformers|"
+                          r"tokenizers|nvidia-.*|cuda-python|cuda-bindings|vllm-flash-attn|flash-attn|"
+                          r"compressed-tensors|llguidance|xgrammar|outlines-core)$")
+
+
+def is_key(row: dict) -> bool:
+    key = row["key"]
+    return key in ("python", "gpus") or key.startswith("env.") or bool(KEY_PACKAGES.match(key))
+
+
+def ranked(rows: list[dict]) -> list[dict]:
+    """Python and GPU first, then key inference packages and env vars, then everything else."""
+    return sorted(rows, key=lambda r: (0 if r["key"] in ("python", "gpus") else 1 if is_key(r) else 2, r["key"]))

servediff-0.1.0.dist-info/METADATA

@@ -0,0 +1,168 @@
+Metadata-Version: 2.4
+Name: servediff
+Version: 0.1.0
+Summary: Diff what your vLLM / SGLang server actually ran with: auto-selected backends, CUDA graphs, startup failures, dependency versions.
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/apex-mochen/servediff
+Project-URL: Issues, https://github.com/apex-mochen/servediff/issues
+Project-URL: Changelog, https://github.com/apex-mochen/servediff/blob/main/CHANGELOG.md
+Keywords: vllm,sglang,llm-inference,performance-regression,debugging
+Classifier: Programming Language :: Python :: 3
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: Topic :: Software Development :: Debuggers
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Provides-Extra: dev
+Requires-Dist: pytest>=8; extra == "dev"
+Dynamic: license-file
+
+# servediff
+
+**Same flags, different server.** `servediff` shows what your vLLM or SGLang server
+*actually* ran with — and what changed between two runs.
+
+Benchmark result files only record the client side. After an upgrade, the server can
+silently pick a different MoE or attention backend, change its CUDA graph capture,
+fall back to another code path, or run against different torch / triton / flashinfer
+builds. Today you find that by reading startup logs and bisecting by hand.
+`servediff` does the reading for you.
+
+- **Who it is for:** anyone comparing two vLLM / SGLang runs — after an upgrade, a flag
+  change, or between machines — who needs to know *what differs on the server side*.
+- **What it reads:** the server's startup log, and optionally a snapshot of the
+  server's Python environment. Offline, no GPU, no dependencies.
+- **What it does not do:** it does not say *which* difference caused a performance
+  change. That still needs an A/B run. It tells you what to A/B.
+
+## Try it in one minute (no GPU)
+
+```bash
+git clone https://github.com/apex-mochen/servediff && cd servediff
+pip install .
+servediff diff examples/issue-excerpts/vllm-47047-v0.23-default.log examples/issue-excerpts/vllm-47047-v0.23-marlin.log
+```
+
+```text
+fact                             baseline                                 candidate                                status
+backend.moe                      VLLM_CUTLASS (NvFp4)                     MARLIN (NvFp4)                           changed
+
+1 difference(s); 0 fact(s) identical. Absent means the log did not print it, not that it was off.
+```
+
+That is the difference the reporter of
+[vllm#47047](https://github.com/vllm-project/vllm/issues/47047) found by hand after a
+v0.21 → v0.23 throughput drop. The example files are the log lines posted in that
+issue (see [tests/fixtures/README.md](https://github.com/apex-mochen/servediff/blob/main/tests/fixtures/README.md)).
+
+## Real example: same flags, vLLM 0.28.0 vs 0.30.0
+
+Two startup logs from one RTX 4060 Laptop under WSL2, identical flags
+(Qwen2.5-0.5B-Instruct, fp16, `--max-model-len 2048`, `--gpu-memory-utilization 0.45`):
+
+```bash
+servediff diff vllm-0.28.0.log vllm-0.30.0.log
+```
+
+```text
+warning: baseline server failed to start: RuntimeError: UVA is not available; facts after the failure are absent because startup stopped, not because they were off
+
+fact                             baseline                                 candidate                                status
+backend.attention                <absent>                                 FLASH_ATTN                               only_candidate
+backend.flash_attn_version       <absent>                                 2                                        only_candidate
+backend.sampling_flashinfer      <absent>                                 disabled                                 only_candidate
+cuda_graph.captures              <absent>                                 ["PIECEWISE x51", "FULL x2", "FULL x35"] only_candidate
+cuda_graph.pool_gib              <absent>                                 0.17                                     only_candidate
+engine.version                   0.28.0                                   0.30.0                                   changed
+startup                          failed                                   ready                                    changed
+startup.error                    ["RuntimeError: UVA is not available", " <absent>                                 only_baseline
+uva_buffers                      <absent>                                 fallback: device memory                  only_candidate
+
+9 difference(s); 8 fact(s) identical. Absent means the log did not print it, not that it was off.
+```
+
+0.28.0 does not start on WSL2 with default settings; 0.30.0 does, because it falls
+back to device memory when pinned memory is unavailable. Upstream fixed this in
+[vllm#56908](https://github.com/vllm-project/vllm/pull/56908), first released in
+v0.30.0. Forcing the V1 model runner on 0.28.0 (`VLLM_USE_V2_MODEL_RUNNER=0`) makes it
+start, and the remaining server-side differences are:
+
+```text
+fact                             baseline                                 candidate                                status
+cuda_graph.captures              ["mixed prefill-decode, PIECEWISE x51",  ["PIECEWISE x51", "FULL x2", "FULL x35"] changed
+cuda_graph.pool_gib              0.36                                     0.17                                     changed
+engine.version                   0.28.0                                   0.30.0                                   changed
+model_runner                     <absent>                                 V2                                       only_candidate
+uva_buffers                      <absent>                                 fallback: device memory                  only_candidate
+
+5 difference(s); 11 fact(s) identical. Absent means the log did not print it, not that it was off.
+```
+
+## Dependency versions: `servediff env`
+
+Logs do not print torch / triton / flashinfer versions, and upgrades of those change
+server behaviour too. Snapshot each server environment with its own interpreter —
+nothing needs to be installed into it:
+
+```bash
+servediff env --python /path/to/old-venv/bin/python -o old.json
+servediff env --python /path/to/new-venv/bin/python -o new.json
+servediff diff old.json new.json
+```
+
+Excerpt from the same two environments (20 inference-relevant rows; 74 other packages
+hidden unless you pass `--all`):
+
+```text
+pkg.flashinfer-python            0.6.16.post3                             0.6.18.post1                             changed
+pkg.nvidia-cuda-nvcc             13.3.73                                  13.0.88                                  changed
+pkg.nvidia-cutlass-dsl           4.6.2                                    4.7.1                                    changed
+pkg.torch                        2.13.0                                   2.13.0+cu130                             changed
+pkg.transformers                 5.16.1                                   5.17.0                                   changed
+pkg.vllm                         0.28.0                                   0.30.0                                   changed
+```
+
+Note the older vLLM environment carries *newer* CUDA toolkit wheels (13.3 vs 13.0).
+Whether that matters for performance is not known — that is what an A/B run is for.
+Versions are read from package metadata, so the same torch build can show as `2.13.0`
+in one environment and `2.13.0+cu130` in another.
+
+The snapshot records Python, GPU name and driver (via `nvidia-smi`), all installed
+package versions, and env vars starting with `VLLM_`, `SGLANG_`, `SGL_`, `CUDA_`,
+`NCCL_`, `TORCH_`, `PYTORCH_`, `FLASHINFER_`, `TRITON_`. Values of variables whose name
+contains `TOKEN`, `KEY`, `SECRET`, `PASSWORD`, `CREDENTIAL` or `AUTH` are replaced with
+`<redacted>`. Snapshots and logs contain local paths: review them before sharing.
+
+## What it extracts
+
+| Fact | vLLM | SGLang |
+|---|---|---|
+| version | banner / engine init | — (not printed) |
+| arguments | `non-default args` | `server_args` (all args) |
+| attention backend | `Using X attention backend` | `attention_backend` arg |
+| MoE backend | three log formats seen in 0.23–0.30 | `moe_runner_backend` arg |
+| CUDA graph | eager mode, `cudagraph_mode`, capture sets, pool size | prefill CUDA graph disabled |
+| model runner, UVA fallback | `Using V2 Model Runner`, pinned-memory fallback | — |
+| startup outcome | ready / failed + root-cause error | ready |
+
+Every fact keeps the log line it came from (`servediff show <log>` or `--json`).
+A fact the log did not print is `<absent>`: servediff never fills in defaults.
+Options shared by both engines (dtype, context length, memory fraction, prefix cache,
+CUDA graph, TP size) are mapped to common names, so a vLLM log and an SGLang log can be
+compared, though the result is noisier than a same-engine diff.
+
+Exit codes: `0` no differences, `1` differences, `2` nothing recognised.
+
+## Limits
+
+- Tested on: log excerpts posted in vllm#47047 (v0.23) and vllm#53472 (v0.23, v0.26);
+  full startup logs of vLLM 0.28.0 and 0.30.0 and one SGLang log, all single-GPU with a
+  dense model. TP > 1, MoE models in full logs, and other versions are not yet verified.
+- Parsing is regex-based on log text, which changes between versions. A missed line
+  shows up as `<absent>`, never as a wrong value — please report it.
+
+## License
+
+Apache-2.0.

servediff-0.1.0.dist-info/RECORD

@@ -0,0 +1,10 @@
+servediff/__init__.py,sha256=BlZahw9q-J1J_yBDzTlFWb8wWKiMt3a9KqYq1MdPvts,189
+servediff/__main__.py,sha256=C9ZV0To-Mbv-Ko-lwbXvJ68zKJF7Z_6UJp-bVeg_6XM,61
+servediff/cli.py,sha256=y6c6d7YSIZruBElWmm6hzJIGyWhQEKobmC_t0DGuc0Q,15061
+servediff/env.py,sha256=0IMgI8GdXcCs_xNm88SCqs66GsAS2KPezGHENjJJGmc,4644
+servediff-0.1.0.dist-info/licenses/LICENSE,sha256=z8d0m5b2O9McPEK1xHG_dWgUBT6EfBDz6wA0F7xSPTA,11358
+servediff-0.1.0.dist-info/METADATA,sha256=haRGvDvaMjpYOtKh8DbVUnILlq8eb6bwvtKDeiil8F0,9483
+servediff-0.1.0.dist-info/WHEEL,sha256=mOdXR9C3wMVov7cSY8PY-XCzo3SmmWCHvGQr5csJoao,91
+servediff-0.1.0.dist-info/entry_points.txt,sha256=FG1eHJxmZNqphvfSEltaumscJyJq0WGR9kqxChbODUI,49
+servediff-0.1.0.dist-info/top_level.txt,sha256=TADHNtBMOkA7WSJj_wjkF7j8Vw558QWox5gTdIhghQs,10
+servediff-0.1.0.dist-info/RECORD,,

servediff-0.1.0.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: setuptools (84.2.1)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

servediff-0.1.0.dist-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+servediff = servediff.cli:main

servediff-0.1.0.dist-info/licenses/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

servediff-0.1.0.dist-info/top_level.txt

@@ -0,0 +1 @@
+servediff